System DietadHoc: A Fusion of Human-Centered Design and Agile

Development for the Explainability of AI Techniques Based on

Clinical and Nutritional Data

Michelangelo Sofo

1,* a

, Giuseppe Labianca

2,* b

, Giancarlo Mauri

3,* c

and

Francesco Combierati

4,* d

IT Consultant, Via Avvocato Vittorio Malcangi, 141/L, 76125, Trani, Italy

Dietician and Nutritionist Biologist, Via Tenente Morrico, 17, 76125, Trani, Italy

University of Milano – Bicocca, Piazza dell’Ateneo Nuovo, 1, 20126, Milano, Italy

Dipartimento di Scienze Economiche, Piazza Scaravilli, 2, 40126, Bologna, Italy

Keywords: Medical Decision Support System, Physiological Data Extraction, Human Centered Design (HCD), Contextual

Design, Participatory Design, Rapid Prototyping, Agile Development, Continuous Optimization Algorithms,

Human Centered Artificial Intelligence (HCAI), Time Series Analysis, Visual Data Mining, Deductive Database.

Abstract: In recent years, the scientific community's interest in the exploratory analysis of biomedical data has increased

exponentially. Considering the field of research of nutritional biologists, the curative process, based on the

analysis of clinical data, is a very delicate operation due to the fact that there are multiple solutions for the

management of pathologies in the food sector (for example can recall intolerances and allergies, management of

cholesterol metabolism, diabetic pathologies, arterial hypertension, up to obesity and breathing and sleep

problems). In this regard, in this research work a system was created capable of evaluating various dietary

regimes for the aforementioned specific patient pathologies. The system is based on a mathematical-numerical

model and is tailored for the real working needs of experts in human nutrition, endocrinologists and cardiologists,

using the Human-Centered Design (HCD - ISO 9241 210). DietAdhoc is a decision support system to the

aforementioned specialists for patients of both sexes (from 18 years of age) developed with an innovative agile

methodology. The software consists in drawing up the biomedical and clinical profile of the specific patient by

applying two implementation approaches on nutritional data.

1 INTRODUCTION

Food science is currently an area of particular

scientist interest, both for software developers, and

for the clinical specialists involved (nutritional

biologist, cardiologists, diabetologists and

endocrinologists). "Do it yourself" diet software,

advertised on internet, television and in newspapers,

are often followed by people with various types of

nutritional problems. In this research work the direct

https://orcid.org/0009-0004-7928-1331

https://orcid.org/0009-0009-3971-5218

https://orcid.org/0000-0003-3520-4022

https://orcid.org/0000-0002-6832-8147

* These authors contributed equally.

dietup.it

easydiet.it

interaction with a dietician - nutritional biologist made

it possible to highlight that the problem with such lies

in the fact that, once tested, they generate approximate

and incomplete results, often causing situations of

malnutrition. Basing on the HCD approach, in the user

– research phase (Mastrangelo et al., 2015), and under

the supervision of the nutritional biologist involved in

the creation of the DietAdhoc system, twelve

commercial software tools (DietUp

, Easy Diet

Sofo, M., Labianca, G., Mauri, G. and Combierati, F.

System DietadHoc: A Fusion of Human-Centered Design and Agile Development for the Explainability of AI Techniques Based on Clinical and Nutritional Data.

DOI: 10.5220/0013054800003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 1, pages 519-528

ISBN: 978-989-758-731-3; ISSN: 2184-4305

519

Evolution Fit

, Fatsecret

, Melarossa

, MetaDieta

Myfitnesspal

, Nutribook

, Nutrium

, Nutriverso

SifaDieta

, Winfood

) have been evaluated in all

their functionality. It was found that the

aforementioned software tools, available through

various subscription formulas, present generic and

redundant functions and therefore are incomplete for

all the clinical. Practically such systems don’t

adequately support end user (specialists in human

nutrition, cardiologists, diabetologists and

endocrinologists) for their lack of explainability and

reliability. During this assessment, it has been

possible to highlight the absence of an exploratory

analysis of a patient’s clinical and nutritional data and

a poor attention to the relevant aspects, specifically in

the following areas:

• allergies and intolerances (shellfish, dried fruit,

mushrooms, gluten, lactose, yeasts, nickel and

other);

• analysis of the pathological history with

possible familiarity (anemia, autoimmune

pathologies, endocrine pathologies, oncological

pathologies and cardiovascular risk);

• functional anamnesis analysis (abdominal pain,

bowel movements, asthenia, menstrual cycle,

dermatitis, dyspepsia, eczema, hiatal hernia,

stools, smoking, gastritis, abdominal swelling,

loss of appetite and myalgia);

• analysis of sports activity (type of discipline and

daily frequency);

• complete evaluation of total blood chemistry

tests (azotemia, blood urea nitrogen, blood

glucose, creatinine, direct bilirubin, indirect

bilirubin, total bilirubin, total cholesterol,

cardiovascular risk index, chemical test of urine,

cholesterol HDL, cholesterol LDL, creatinine,

eGFR, mGFR, complete blood count, ferritin,

glycolysed hemoglobin (HbA1c), Homa Index

IR, Homa – β, insulinemia, homocysteine,

hormonal dosages, iron level in blood,

triglyceridemia, transaminases GOT,

transaminanes GPT);

• any therapies in progress, or taken in the last few

months (name of the drug with relative dosage);

• section in which the nutritional biologist can

insert complementary clinical notes;

• dynamic configurators for determining the

macronutrients to be consumed daily

evolutionfit.it

fatsecret.it

melarossa.it

metadieta.it

myfitnesspal.com

(carbohydrates, proteins and fats) for evaluating

the kilocalories and micronutrients (cholesterol,

fibers, potassium, sodium, etc.) to be consumed

in the various meals of the day;

• ORAC values (antioxidant power of foods) to

counteract cellular aging (Haytowitz D.B. et

al.,2010);

• PRAL values (renal acid load potential) to

calculate the chemical balance of the acidifying

and alcalinisant molecules of a food (Remer et

al., 1995);

• BIA data, (obtained from the bioelectric

impedance analysis) – (Mehra et al., 2024);

• glycemic index of a food (Scazzina F. et al.,

2016);

• glycemic load, calculated based on the quantity

of carbohydrates in the food portion in grams for

the specific glycemic index (Scazzina F. et.al.,

2016);

• absence of the glycemic curve of the meal;

• absence of the weight history (current weight,

desirable weight, ideal weight).

These gaps obviously cause an approximation of the

patient's clinical profile. As will be explained in detail

in the following chapters, the objective of this research

word consists in creating a virtuous integration

between the consolidated clinical experience of a

specialist in human nutrition and its supporting

technical equipment to create a software that improve

patient well-being and help prevent diseases (Figure

1).

Figure 1: DietAdhoc’s virtuos integration.

nutribook.app

nutrium.com

nutriverso.cloud

sifadieta.com

winfood.it

BIOINFORMATICS 2025 - 16th International Conference on Bioinformatics Models, Methods and Algorithms

520

2 DIETADHOC DESIGN

During the user research phase of HCD approach, the

designer of the DietAdhoc system takes into account

various clinical checks and visits carried out by the

nutritional biologist (ethnographic research).

Following this process, the system designer develops

a series of unstructured interviews and specific

questionnaires to define:

• the complete profile of the characteristics of the

nutritional biologist and of the designer –

system administrator who will use the system

[age, gender, type of interaction with the system

(primary or secondary), frequency of use of the

system (daily, in the specific case), level of

computer experience (inexperienced, sufficient,

good, expert), system usage time and technical

support equipment];

• task analysis (both for the nutritional biologist,

and for the designer - system administrator);

• usage scenarios (both for the nutritional

biologist, and for the designer – system

administrator);

• future stakeholders (cardiologists,

diabetologists and endocrinologists), interested

in scientific collaborations to share clinical

information for the creation of a

multidisciplinary team).

In the initial phase, the two figures in charge of

the software will be the nutritional biologist (domain

expert) and the system designer who will also act as

system administrator. The nutritional biologist, is 32

years old and has basic computer skills. He represents

the primary user of the system because he uses it daily

and deals with the insertion and update of all the

patient’s clinical data (including reports from other

specialists), of a new food portions relating to certain

diets into the database, and for the generation of

clinical reports to be sent via e-mail to patients). The

designer – system administrator, is 42 years old, and

has a high level of computer skills. He represents the

secondary user of the system because he is contacted

on the base of the needs of the nutritional biologist.

The system designer remotely checks for any

malfunctions in the workflow and in the interfacing

with the technical instrumentation.

Subsequently the nutritional biologist has

contacted other medical specialists (cardiologists,

diabetologists and endocrinologists) for the creation

of a scientific multidisciplinary team to support the

DietAdhoc® system. The current technical

instrumentation supporting the human nutrition

specialist consists of: SECA 799 digital column scale,

the SECA mBCA 525 digital bioimpedance meter and

the GIMA digital caliper. At the moment the system

administrator is in contact with the commercial and

technical managers of the aforementioned medical

devices to evaluate the possibility of inserting the

DietAdhoc system into them. Finally, based on the

data obtained from unstructured interviews and

questionnaires, also addressed to the aforementioned

clinical specialists, the designer created usage

scenarios, i.e. descriptions in natural language, of how

the application (with all the tasks identified) will be

used by current and potential users in order to verify

all the needs expressed.

2.1 Requirements

It was found that a software to support a professional

in the field of nutrition, compared to classic data

processing applications coded in a procedural manner,

had to manipulate a knowledge of a clinical nature

with an inferential approach and had to allow an

analysis of the patient with three levels of detail:

• descriptive – analyze the data of the various

patients to decipher the latent details that escape

medical procedures;

• predictive – creation of analytical and complete

models from the aforementioned latent data with

the aim of predicting valid results from the

scientific community in the nutritional field;

• prescriptive - suggest corrective clinical actions

for a given patient, using all the information that

the system has generated.

In this regard, the operational knowledge was codified

in the system by the designer and the following high

fidelity prototypes were created (Conceptual design -

second phase of HCD process):

• a section concerning four food categories (main

food, vegetables, fruit and drinks) so that the

nutritional biologist can insert, update and delete

food portions with the relative contribution in

kilocalories (kcal), macronutrients

(carbohydrates, proteins and fats) and other

nutritional parameters (cholesterol, ORAC,

PRAL, and salts);

• a section for all patient data (name, surname, age,

birth of date, gender, identity document, tax id

code, telephone reference, e-mail, working

activity, eventual Vat number);

• a section for entering medical reports of other

professionals;

• a dynamic menu for evaluating the patient's

medical history, blood chemistry tests and

System DietadHoc: A Fusion of Human-Centered Design and Agile Development for the Explainability of AI Techniques Based on Clinical

and Nutritional Data

521

current therapies;

• an accurate generator of anthropometric

parameters based on seven basic measures

from which all the others are derived;

• two menus for evaluating bioelectric

impedance vector analysis (BIVA) and the

analysis with digital caliper;

• two automatic food regime configurators;

• three heuristic algorithms based on clinical

and nutritional data.

In the third phase of HCD process (Evaluation),

the nutrition biologist carries out a thinking aloud

usability test, designed by the system administrator,

to evaluate the prototypes created in the conceptual

design. All the functions can be viewed in the

demonstration session of the system present at this

link:

https://drive.google.com/file/d/1tnHvpCYAw444Oc2

8-mighNLX9SmV_Of-/view?usp=sharing

From the aforementioned demo, it has been found

that the HCD design process is the conceptually

correct model for the creation of decision support

systems for human health professionals; the product

is seen, even partially, from the beginning of the

process and is perfected in subsequent increments. In

this way, the nutritional biologist, continuously

interfacing with the system administrator, can

continually experiment with each implementation

choice, immediately discarding the wrong ones

(Figure 2).

Figure 2: Iterative HCD Design Process adopted for

DietAdhoc System.

Each prototypes of the DietAdhoc® system have

been developed in Java language with an agile

methodology (Figure 3) until the beta version of the

DietAdhoc system is created.

Figure 3: Agile development adopted for DietAdhoc system.

The iterative design, the user involvement, the

continuous prototyping and testing unite HCD Design

process and Agile Development.

3 MATHEMATICAL MODEL

Unlike commercial software in the biomedical-

nutritional field, mentioned in the section 1, the

DietAdhoc system is founded on a mathematical

model which provides scientific evidence and

empirical evidence to this research work. The model

determines an optimal requirement of daily

kilocalories (in kcal), according to certain constraints

on the requirement of macronutrients (carbohydrates,

proteins and fats in grams), micronutrients

(cholesterol, fibers, potassium, sodium, etc.) and other

nutritional parameters (ORAC, PRAL, glycemic

loads, and salts) basing on patient’s clinical situation.

The specific dietary regime is automated by the

system DietAdhoc which supports the professional in

reaching the final diagnosis through a process of

successive steps with the intervention of the doctor (as

supervision). This model is based on a problem of

continue optimization, therefore the decisional

variables (the kilocalories of each portion of food

belonging to four food categories) are defined on R

(continue values). The scope of the study consists to

maximize a quantity (the daily kilocalories for the

specific diet, given by the sum of each combination of

four portions of food, or of n portions chosen through

the next two algorithms) through a function f: R

→ R

with S ⊆ R

according to the notation max{f(x): x ∈

S}. The function f(x) is a function of n real values

f(x

,…,x

), called objective function, and the set S,

the admissible set, i.e. the set of possible solutions to

the problem. The set S is a subset of ℝ

, so x =

,…, x

)

is an n-dimensional vector variable.

Every x ∈ S is an admissible solution. The admissible

BIOINFORMATICS 2025 - 16th International Conference on Bioinformatics Models, Methods and Algorithms

522

set S is described by a finite set of inequalities, or

equations, of the type g(x) ≥ b, g(x) = b, g(x) ≤ b

where g being a real-valued function defined on ℝ

while b ∈ ℝ. More formally, assigned m functions g

ℝ

→ ℝ, i = 1,2…, m. and m real scalars b

, we have

that S is expressed in the form: S = {x ∈ ℝ

| g

(x) ≥

, i =1,2,…m}. All of that inequalities g

(x) ≥ b

represents a constraint, (the daily grams of nutrients

to be consumed for the specific diet) and the

admissible set is therefore formed by all those points

x which are solutions of the system of inequalities.

More specifically the objective function f(x) and all

the functions that define the constraints g

(x), i =

1,2…,m. are linear, that is expressible in the form

+…+c

(Figure 4) where we may take the

costs ci to be equal to unity (c=1) since in a clinical

treatment it is not important to consider the economic

side of food portion.

Formally, by introducing c ∈ ℝ

(cost vector),

defined as c = (c

,…, c

)

, x ∈ ℝ

, and x = (x

,…,x

)

the objective function can be written in vector

notation as:

Figure 4: Objective function.

The decision problem consists in determining the

portions (in kcal) of each portions of food, x

, j =

1,2…,n, for the particular diet to be followed, so that

the quantity of the j-th nutrient, present in a portion of

food, is greater than, or equal to, the recommended

value bi, daily requirement (in grams) of the j-th

nutrient.

4 SYSTEM DESCRIPTION

In the specific clinical case, after entering of all the

patient’s clinical data and with the automatic

generation phase of anthropometric values (phases

that can be viewed in the demonstration session of

software in section 2.1), the nutritionist biologist will

evaluate a first dietary regime for the particular diet.

The nutritionist biologist can set the dynamic

configurator that determines the kilocalories of the

diet (1413 – the objective function distributed for four

meals of the day) and the quantity in kilocalories, and

in grams, of macronutrients to be consumed.

Figure 5: Diet configurator.

Two computational approaches have been applied

to process the clinical and nutritional data generated

by the system. The first approach, based on the

enumerative - heuristic algorithm with constraints,

has been designed according to three levels of

abstraction:

• planning of the specific meal which

corresponds to the choice of a combination

made up of four portions of foods belonging to

the categories (main food, vegetables, fruit and

drinks), to be taken within the specific meal

(breakfast, morning snack, lunch, afternoon

snack , dinner and third snack);

• planning of the day which corresponds to the

choice of combinations made up of four

portions of foods to be eaten over the course of

a day (breakfast, morning snack, lunch,

afternoon snack, dinner, and evening snack);

• planning of the week which corresponds to the

choice of the daily combinations made up of

portions of four foods to be consumed over the

course of a week.

This algorithmic approach is based on the first

specialist's modus operandi and considers a

decomposition of the original problem into simpler

sub-problems to solve. It is an implicit enumerative

procedure in several steps. Initially it generates all the

possible solutions (combination of four portions of

food in the research space), then in the next steps the

specialist evaluates the heuristic (“MEDIA

CHILOCALORIE”) and sets the bounds to choose

the optimal food combination of four meals for the

specific patient based on his specific eating disorder.

Naturally, for the specific nutritional problem, the

operating logic of the step of choosing the optimal

food combination is controlled by the nutritional

biologist and is extended for the six meals of the day.

From a conceptual point of view, the technique

implemented in the system for the first enumerative -

System DietadHoc: A Fusion of Human-Centered Design and Agile Development for the Explainability of AI Techniques Based on Clinical

and Nutritional Data

523

heuristic algorithm adopts the following steps:

1. Automatic generation of the n-ary tree of the all set

of eligible solutions S (combination of four portions

of food in the research space) based on the value of

the objective function (kilocalories for the specific

meal – i.e 565 kcal for breakfast) set by the nutritional

biologist through the diet configurator (Figure 5);

2. First summary evaluation by the nutritional

biologist of the n admissible solutions (n child nodes

of the n-ary tree) based on the objective function (565

kcal for breakfast) and average value (of all food

combination) 312 kcal like heuristic, of all food

combinations (Figure 6);

3. The nutritional biologist sets lower and upper

bounds (operation on the objective function) to avoid

the evaluation of all food combinations of the four

food portions (method not feasible from a

computational point of view) (Figure 7);

4. Evaluation of the optimal feasible solution, f(x*),

in the range of lower (i.e. 396 kcal) and upper (i.e.

442 kcal) bounds of the objective functions and

completed by an integrated step based on a local

search approach that explores a neighborhood N: S

→ N(S) subset of the feasible set S, and determined

by the bounds and the heuristic, the average of food

combinations included in the bounds (i.e. 415 kcal).

In choosing the most suitable food combination for

the specific meal of the day, the nutritional biologist

will be able to take advantage of the clinical

information (histogram of nutrients and glycemic

curve of the combinations) present in the sections

“NOTE CLINICHE” and “STATISTICHE” (to be

viewed in the software demo session – section 2.1) to

evaluate the constraints (cholesterol, fibers,

potassium, sodium, etc.) of each combination of four

portion of food (Figure 7).

5. The food combination chosen for breakfast by the

nutritional biologist is the following: gran cereale (4

biscotti – 100 g) , fiocchi di avena (30 g), banana (25

g) and yoghurt greco liquido (150 g);

6. Iteration of the previous steps for all remaining

meals of the day determined through the diet

configurator (lunch, afternoon snack, and dinner –

Figure 5);

7. Stop criterion (the nutritional biologist chose the

daily food combination).

Note 1 - For the first computational approach, the

"SALTA PASTO" option was programmed which

allows not to consider a combination of food portions

for each of the six programmed meals (breakfast,

morning snack, lunch, afternoon snack, dinner,

evening snack).

Figure 6: Running demo session of the steps 1 and 2 of the

enumerative - heuristic algorithm with constraints.

Figure 7: Running demo session of the steps 3 and 4 of the

enumerative - heuristic algorithm with constraints.

The first heuristic algorithm has been

implemented through a two-dimensional array of real

numbers, double[][] foodmatrix = new double [n][m];

having as rows, n, the portions of the food for the

specific meal to which the relevant kilocalories are

associated, while as the m columns (m=5) the specific

food category (main, food, vegetables, fruit and

drinks) and like last columns the sum of the

kilocalories of the selected food combination. The

second solution approach, based on a combinatorial -

heuristic algorithm with constraints, has been

designed according to three levels of abstraction:

• specific meal planning - corresponds to the

choice of n portions of foods, belonging to the

categories (main food, vegetables, fruit and

drinks), to be taken within a specific meal

(breakfast, morning snack, lunch, afternoon

snack, dinner, and third snack), calibrated for a

single meal;

• planning of the day - corresponds to the choice

of n portions of foods, belonging to the

categories (main food, vegetables, fruit and

drinks), to be taken within the six specific meals

BIOINFORMATICS 2025 - 16th International Conference on Bioinformatics Models, Methods and Algorithms

524

(breakfast, morning snack, lunch, afternoon

snack, dinner and third snack), calibrated for a

day;

• planning of the week - corresponds to the choice

of n portions of foods, belonging to the

categories (main food, vegetables, fruit and

drinks), to be taken within the six specific meals

(breakfast, morning snack, lunch, afternoon

snack, dinner and third snack), calibrated for a

week.

The idea behind the second algorithmic approach

is to adopt an expansion criterion based on the most

"promising" decision in a given step of the problem

to reach an optimal solution for the given meal of the

day, f(x*), compatibly with the constraints imposed

on the nutrients to be respected for the specific diet.

The expansion criterion is updated iteratively to take

into account previously evaluated decisions. The

combinatorial algorithm with constraints created

allows the nutritional biologist to iteratively add, or

not, the portions of food with relative kilocalories

until arriving at a complete solution (expansion

criterion for the specific meal of the day). At each

iteration, the portions of food, with relative

kilocalories, that produces the greatest improvement

in the objective function for the specific meal of the

day is added (automatically determined by the food

regime configurator under the supervision of the

nutritional biologist). This algorithmic approach is

applicable if the solution (a combination of n food

portions for all meals of the day) can be obtained as a

subset of input data (all the possible combination of

m food portions for all meals of the day) – (Figure 8)

Below is the formalization of the data of the second

algorithmic approach:

S.P. = Specific problem (maximization of the

objective function based on kilocalories of food

portions for all meals of the day), established by the

nutritional biologist and evaluated by the diet

configurator with the relative constraints imposed on

nutrients;

S = Set of eligible solutions (food portions for the

specific daily meal: breakfast, morning snack, lunch,

afternoon snack , dinner and evening snack);

optimal

= Subset of S, consisting of optimal portions

of food chosen by the nutritional biologist for the

particular diet;

partial

= Partial solution for the specific meal of the

day (breakfast, morning snack, lunch, afternoon

snack, dinner, evening snack) chosen by the

nutritional biologist for the particular diet;

Beyond the conceptual scheme of the second

algorithmic approach is the following:

1. Initialize the problem instance S.P.;

2. Evaluate the set of admissible solutions S;

3. For each choice to be made, the nutritional biologist

makes the optimal decisions respecting the nutritional

constraints of the specific problem (S.P.);

4. The criterion for inserting the n portions is

dynamically updated in order to take into account the

choices made previously to determine the partial

solution S

partial

There are two stopping criteria:

1. predetermined number of choices, made by the

biologist nutritionist, for the specific daily meal;

2. obtaining final optimal solution consisting of most

suitable portions of food chosen by the nutritional

biologist for all daily meals.

Beyond the conceptual scheme, it is possible to

formalize the combinatorial- algorithm with

constraints in pseudocode:

INPUT S.P. (Instance of the specific problem)

S ← S

partial

(Initializing partial solution for S.P.)

WHILE S can be extended DO

Find the S

optimal

extension of S: S ← S

optimal

END WHILE

OUTPUT S

Figure 8: Running demo session of the combinatorial –

heuristic algorithm with constraints.

In the specific case:

S.P. = maximization of the kilocalorie value

established by the diet configurator for the specific

meal (565 kcal for breakfast)

S = totality of the specific clinical choices of the

nutritional biologist based on the food portions

foreseen for the specific meal (breakfast);

optimal

= choice by the nutritional biologist of all the

food portions, among the various available , most

suitable for the specific daily meal and for the

particular diet;

System DietadHoc: A Fusion of Human-Centered Design and Agile Development for the Explainability of AI Techniques Based on Clinical

and Nutritional Data

525

partial

= gran cereale (4 biscotti – 100 g), budino

vegano alla vaniglia (100 g), banana (25 g) and caffè

con dolcificante e kefir bianco (150 g).

The algorithm was implemented through a

circular queue. In this way the nutritional biologist

can eventually delete [dequeue operation –

Dequeue()] the choice of food portions belonging to

four categories (main food, vegetables, fruit and

drinks). Seeing the demonstration session present in

paragraph 2.1, the nutritional biologist has a real time

evaluation of the update, or modification of the

objective function (kilocalories associated with food

portions for the specific meal) – (Figure 8).

5 DEDUCTIVE DATABASE

In the previous chapters, two heuristic algorithms

were shown that allow the nutritional biologist to

examine various food combinations and determine

the most suitable ones for the specific patient. At the

same time, the DietAdhoc system can have a

symbolic approach. If Prolog programs are

constrained to use only atoms, integers, and reals, and

do not allow recursive rules, you get a powerful

subset of SQL. Under these assumptions, Prolog and

SQL share a core: every query expressible in a subset

of Prolog can in turn be expressed in a subset of SQL,

that means these subsets are logically equivalent

(Warren D.S., 1999).

For this research work, it has been studied the

possibility of interfacing SWI - Prolog with the

DBMS at the basis of the DietAdhoc system through

the ODBC driver (Figure 9), transforming the

relational database underlying the system into a

deductive database. Prolog rules has been created

which allow the connection with the relational

database of the DietAdhoc system, access its tables,

carry out queries (even nested), generate all the

anthropometric parameters of the patient and evaluate

the enumerative - heuristic algorithm (Figure 10).

Figure 9: Connection with the deductive database.

Figure 10: Enumerative - heuristic algorithm in the form of

Prolog rules of a deductive database (corresponding to the

specific Java interface).

6 EXPERIMENTAL RESULTS

For the first and the second algorithmic approaches,

once the nutritional biologist chooses the food

combinations for the particular diet, the system

generates a final daily report in which it is possible to

view all the data obtained with flexible and

customizable user interfaces for the purposes of the

various clinical assessments (Figure 11).

Figure 11: Decision support through various graphical

components.

The nutritional biologist will be able to view the

composition of the macronutrients and micronutrients

of the chosen food combinations via a special

histogram (Figure 11). In the case of the glycemic

curve (Carico glicemico) for the diagnosis of diabetes,

through a time series the Y phenomenon was modeled

through five observations at time t, corresponding to

the six daily meals, with integer t varying from 1 to t.

(Figure 11).

The vector bioelectrical impedance analysis

(BIVA) is a methodology developed in the second half

of the 1990s, which uses vector models and is based

on the electrical properties of tissues without the use

BIOINFORMATICS 2025 - 16th International Conference on Bioinformatics Models, Methods and Algorithms

526

of constants, equations and body weight. These

characteristics, together with the low cost and speed

of execution, make BIVA the type of bioimpedance

analysis most used for all those patients who present

alterations in renal and/or cardiac function, or who

find themselves in conditions of extreme

malnutrition, as well as for cancer and neurodeprived

patients, who may have difficulty interacting

correctly or enduring exams that take too long

(Campa F. et al., 2023). The report includes a

graphical representation (vectors) of the body

resistance and reactance values (Figure 12). This

therefore allows an evaluation of the subject's

hydration and nutrition status independently and

without a mathematical calculation of the data based

on the resistance/reactance detected.

Figure 12: BIVA data.

7 COPYRIGHT

The algorithms implemented and the source code of

the DietAdhoc system are protected by copyright law

(Legislative Decree 518/1192, E.C. Directive 250/91,

Law 747/1994, L.633/1941 Articles 20 and 24).

Anyone who violates the aforementioned regulations

faces civil and criminal sanctions relating to the legal

protection of the DietAdhoc software (Law

248/2000).

8 CONCLUSIONS

Basing on the initial clinical objectives agreed with

the domain expert, it has been demonstrated that two

computational approaches, enhanced by the complete

explanation of the decisions generated, can configure

diets supported by the scientific community

providing an aid to the diagnosis of various eating

disorder with the generation of a complete prediction

model.

Some system upgrades are currently in development:

• Import and integration of biomedical data of a

semi-structured nature (json and xml), and in csv

format into the system DietAdhoc;

• Distributed evolution of software for the creation

of a collaboration network between experts in the

nutrition sector and other specialists;

• Human – Centred AI (HCAI) techniques which

allow to have a complete vision of all the data

generated by the system for the purpose of

clinical choices;

• Insertion of predictive machine learning

algorithms for the exploratory analysis of clinical

and nutritional data;

• Symbiotic Artificial Intelligence (SAI)

techniques for support, without replacing them,

other specialists in human nutrition, and

professionals in the medical field interested in a

scientific collaboration;

• Visual Data Mining techniques to discover

regularities and patterns in patients with similar

ongoing therapies;

In subsequent papers, a framework will be

proposed to accelerate the transfer of artificial

intelligence to clinical contexts and to personalize the

treatment of all nutritional pathologies

REFERENCES

Mastrangelo S., and Boscarol M. (2015) – Linee guida per

la progettazione centrata sull’utente (Versione beta 0.1)

– Gruppo di lavoro per l’usabilità - https://

www.funzionepubblica.gov.it/sites/funzionepubblica.g

ov.it/files/linee_guida_appalti_hcd_beta01_0.pdf

Bhagwat S., Haytowitz D.B and (2010) – USDA Database

for the Oxygen Radical Absorbance Capacity (ORAC)

of selected food – Nutrient Data Laboratory, Beltsville

Human Nutrition Research Center (BHNRC) –

Agricultural Research Service (ARS) - U.S. Department

of Agriculture (USDA) MD May 2010 -

https://www.ars.usda.gov/

ARSUserFiles/80400525/articles/airc07_orac.pdf

Remer and Manz F. – (1995) - Potential renal acid load of

foods and its influence on urine pH - Journal of the

American Dietetic Association Volume 95, Issue 7, July

1995, Pages 791-797 - https://doi.org/10.1016/S0002-

8223(95)00219-7

Mehra A, Starkoff B.E. and Nickerson B. S. (2025) – The

evolution of bioimpedance analysis – Nutrition Volume

129, January 2025, 112601 -

https://doi.org/10.1016/j.nut.2024.112601

Scazzina F., Dall’Asta M. and Casiraghi M.C. (2016) -

Glycemic index and glycemic load of commercial Italian

System DietadHoc: A Fusion of Human-Centered Design and Agile Development for the Explainability of AI Techniques Based on Clinical

and Nutritional Data

527

foods - Nutrition, Metabolism and Cardiovascular

Diseases Volume 26, Issue 5, May 2016, Pages 419-429

– https://doi.org/10.1016/j.numecd.2016.02.013

Warren D.S. (1999) - Prolog as database query language –

Introduction to Prolog (section: deductive database)

https://www3.cs.stonybrook.edu/~warren/xsbbook/nod

e11.html#:~:text=Prolog%20is%20an%20elegant%20l

anguage,a%20powerful%20subset%20of%20SQL.

Campa F., Coratella G., Cerullo G., Stagi S., Paoli S. Marini

S., and Grigoletto A. (2023) - New bioelectrical

impedance vector references and phase angle centile

curves in 4,367 adults: The need for an urgent update

after 30 years https://doi.org/10.1016/j.clnu.2023.

07.025 - Clinical nutrition Volume 42, Issue 9,

September 2023, pages 1749-1758

BIOINFORMATICS 2025 - 16th International Conference on Bioinformatics Models, Methods and Algorithms

528